Valuation Anchoring

They say one should start an article with a memorable line. So, how about this: In FY26, SK Hynix is on track to generate more profit than Nvidia. 

Of course, there are qualifiers. NVIDIA’s fiscal year ends in January, while Hynix wraps up in December. For the calendar year that constitutes most of 2026, Nvidia likely still holds the lead. But, arguing over the calendar is like debating the font on a winning lottery ticket, particularly when one of them making over USD100bn in profits trades at less than five times its earnings. 

We are told that this is the valuation fate of being a commodity company. Let’s ignore the fact that what should be called a commodity should be traded in open markets and not over multi-year supply contracts or be available freely, even if fortuitously, to some. You cannot find a profitable cotton company or a profitable coffee company in any emerging market trading at that kind of multiple, let alone a company whose products cannot be matched easily by the closest competitors and sold directly to only a small number of buyers whose own highly profitable existence depends on ensuring that the supply does not go to their competitors.

The Korean discount makes an appearance. Never mind that there is no dearth of technology and pharmaceutical companies, some even from the same group, trading at multiples themselves 5 or more times higher. 

Historical price-to-book is quoted with the seriousness of scripture. Even if one were to look askance at the new ROE peaks at 70%, the detractors do not feel the need to take a few years forward book values discounted back for the right multiple, despite the book value growth of 50 percent a year this year and in excess of 30% cagr for the next few (for emphasis: 30%+ expected cagr in the book value).

There is a stranger dimension to this immunity. When earnings estimates are revised not by a few percentage points but by a few multiples inside a single year, the stock develops a curious numbness to the magnitude of the upgrade. Whether the completely unexpected long-term contracts were for 1 or 3 instead of 5 years, market reactions would not have been different. Memory prices for the quarter could have risen by 10% or 30% instead of 50% or 70%, and the target prices would not have changed.

The fate is no different for the listed memory maker, Micron, in the US, which pours cold water on the revaluation hypothesis built around the ADR listing. 

Looking at these stocks’ gyrations, there must be more sense in optical names trading at fifty times forward earnings despite every segment attracting more competition by the day or even those making capacitors or power ICs, all tagged as commodities and without any multi-year contracts or similar innovation moats. 

Valuation anchors are more difficult to kill than fax machines in a Tokyo trading desk. Hynix, Samsung or Micron are unlikely to make an appearance as a case study in any book promoting efficient markets.

Macro Tagging

We love a good label. Let’s call it macro tagging. 

Finance has many gifts, but perhaps none more beloved than the ability to compress a complex business into a single, tradeable handle. A Korean chipmaker is not a company with multi-year supply contracts, proprietary stacking architecture, and the singular distinction of being the only credible supplier to the world's most valuable semiconductor ecosystem. It is an EM beta play. A Chinese biotech with a best-in-class oncology pipeline and peer-reviewed trial data is not an innovator. It is a geopolitical risk proxy. Brazilian fintech building the most sophisticated credit scoring infrastructure in Latin America? A real rate bet with a user interface.

The habit is understandable. Macro moves fast and arrives without an appointment. Risk managers need something to point at when the book swings thirty percent in a week, and "the business is fine, actually" is not a satisfying answer at four in the morning. A macro tag is efficient. It explains everything and requires understanding nothing.

The problem is not the tag itself. The problem is the half-life. Macro events pass. Trade wars resolve, or mutate into something unrecognisable. Rate cycles turn. Elections conclude, even if the arguments do not. The tag, however, stays. Long after the original macro thesis has expired, the label clings to the stock like a barnacle to a ship that has already crossed the ocean and docked somewhere entirely different. Even if the last macro events are over, the biggest certainty is that there will be the next ones where these tags will again have use.

And so the stock becomes curiously immune to its own news. Contract wins, margin expansion, a competitor stumbling: none of it moves the needle quite like a central bank press conference or a tweet from a trade minister who may not even be in office by Thursday. The business keeps compounding quietly in the background while the front page decides the price.

There are actors who drive the prices, and there are performers with opinions about how the current macro will make the markets move in some inevitable way for the periods to come. For the incorrigible bear, each is a harbinger of the deeper crash that is surely coming this time. For the incorrigible bull, every recovery is evidence that fundamentals always win, conveniently forgetting that they spent the drawdown as rattled as everyone else.

In the crossfire of these competing certainties, the investor trying to follow an actual business is left with the least glamorous of all strategies: twiddle the thumbs. There is theoretical justification in the stubborn insistence that footnotes matter more than headlines. That is what the efficient markets books say. But did we not say something about those books already?

Markets are a combination of those investing top-down and those doing it bottom-up. Few manage both well. In any short stretch, the top-down dominates: loudly, confidently, and with excellent television graphics. For those investing in businesses and innovation, there is nothing but the quiet hope that in the long run, it is not entirely about everyone being dead.

Capex Recoiling

Macro-wise, the tech world changed around 2022. Some of the changes are more accepted than before, but the one most difficult to digest is the end of successful tech's most cherished privilege: making serious money without investing anything serious first.

This was not always considered a privilege. Across every other industry in every other era, you had to buy the ship before you could trade the spice. Capital expenditure was not a pathology. It was simply the price of ambition. Then came thirty years of software, which convinced an entire generation of investors that growth was something you achieved without the inconvenience of building anything physical at all. The reflex hardened into dogma. A large investment plan became read less as ambition and more as an admission that something fundamental was missing. Capex was a confession.

The phobia is now misfiring at considerable cost. In skeptical markets, particularly across Asia and most acutely in China, the sell-side machinery revolts at the sight of a heavy capex line. Analysts who will cheerfully model ten years of speculative software revenue will downgrade a company for building even a cloud business. In the US, large capital outlays are the pessimists’ biggest weapon. The new breed of fundamental analysts, who never bothered about cashflow statements for almost any company of any kind, will have instant concern about even falling free cash flow, never mind the tens of billions sitting in treasury earning low single digits. Apple, one of the most celebrated cash hoarders in corporate history, attracts elaborate theories about how not attempting anything might be the winning strategy. 

We are now in an era where the moat is often the spending itself. TSMC spends forty billion dollars a year so that no one else can. NVIDIA's supply chain investments lock out competitors years in advance. The hyperscalers are building data centers like they expect the world to run on them. Each of these companies is treated with the suspicion that would have been appropriate for a dot-com-era furniture retailer building its own warehouses.

The market looks at a hundred-billion-dollar build-out and sees a suicide note. In reality, it is the highest form of competitive violence. The moat is no longer about clever algorithms or charismatic founders. It is the physical reality of having the chips, the power, and the cooling when no one else does. To dominate the digital future, you must first own the physical present. The real power is shifting to the basement where the servers actually hum, but the investors remain upstairs, admiring the interface.

And here the macro loop closes. Infrastructure at this scale cannot be staged or financed in tranches. It makes the new technology era structurally more dependent on rates, credit availability, and sovereign appetite than the idea-based era ever was. The very innovation investors, like us, who don’t want to think about macro as much as the rest are more exposed to macro forces than ever before.

Aura Pricing

Every industry builds temples for its heroes. Technology simply replaces stained glass with keynote stages and VIP greenrooms. The habit made sense when software could be willed into existence by a handful of brilliant coders in a rented house. But today's most consequential innovations are giants and institutional. The physics of advanced packaging or sovereign compute cannot be brute-forced by sheer charisma. Yet celebrated names from the last cycle command billion-dollar valuations on the strength of a sketched napkin, while the market assigns punishing discounts to companies with impenetrable moats simply because their executives do not host podcasts.

The conference circuit is where this becomes most visible. There is an entire economy built around the speaker slot, the dinner invitation, the proximity that signals access. The sell-side knows which names fill rooms. Allocators know which founders make the pitch worth taking. The actual business, the one with contracts and margins and a credible competitor breathing down its neck, becomes almost incidental to the persona projected on stage. An analyst note will spend three paragraphs on a founder's vision and one sentence on the product. The market pays a premium for the story and discounts the operator. We buy the shirt because of the logo. We ignore the fabric.

When the biggest names, like Elon Musk and Zuckerberg in the field of the LLM development, publicly struggle with the very problems their reputations were built on, the cult adjusts its theology rather than questions its faith. The timeline has just shifted. The capital destruction was a necessary sacrifice. It is easy to mock the retail trader buying a ticker because of a late-night social media post. It is much harder to admit how many institutional investment committees wave through a mediocre thesis because the founder has a great aura.

The new era demands a different kind of leader entirely. Navigating a highly disruptive, explosively evolving environment requires the humility to accept that nothing is given about where things are headed. It requires a different humility still with partners and vendors who could once be squeezed on price: the willingness to visit their sites, sit down for fried chicken dinners, and treat supply chain relationships as strategic assets rather than negotiating targets.

In the capex-heavy era, personality may still play a disproportionate part in luring cheap capital. That may be its last remaining superpower.

Curvilinear Thinking

When Sora launched, the investment world mapped out a five-year global transformation before the first user had even produced a believable cat video. We habitually assume the launch curve is the product curve. It is a comforting fiction. We take a moment of genuine awe and stretch it into a multi-year revenue model, which is the financial equivalent of booking a victory parade after the first successful rehearsal.

The habit treats all innovation as travelling at a single, uniform speed. It misses the structural speed limits. HBM moves with the predictable cadence of a heavy machine because the next upgrade is locked into a multi-year roadmap controlled by a tiny group of players with forty-billion-dollar factories. Optical transceivers exist in a state of permanent fever. The 400G cycle had barely finished its coffee before 800G kicked the door down. Now 1.6T is already in qualification, and transitional architectures like LRO are being declared obsolete before they have cleared the warehouse. Same industry, completely different clock speed. The model assumes one; nature provides many.

The deeper problem is that a DeepSeek or a Claude Code moment can now arrive on an annual basis in multiple innovation fields and render a three-year competitive moat decorative in a jiffy. Software switching costs, proprietary data advantages, incumbent distribution — each of these was a durable moat until the cycle decided otherwise, often without warning and certainly without consulting the discounted cash flow model. Terminal value calculations rest on technological paradigms that are being renegotiated every few quarters. A five-year steady-state margin assumption in this environment is not analysis. It is fan fiction with a discount rate attached.

The honest version of the multi-year model would include a column labelled "probability that something we cannot currently name makes this irrelevant." It would be the widest column on the page.

We know this. We write about it. We have published pieces on precisely this acceleration. Then the quarterly earnings season arrives, and we find ourselves debating the fair multiple based on 2028 earnings.

General Grievance

We have a deep, almost religious affection for the specialist. In a world of overwhelming complexity, we want the person who has spent thirty years looking at a single type of bolt or making a specific kind of sushi. We call it domain expertise. We treat it as the ultimate defense against the amateur. We believe that depth is the only true measure of brilliance and intelligence.

We carried this prejudice into the AI era like a favorite old coat. We convinced ourselves that Vertical AI was the only path to a durable moat. We wanted the model that did nothing but read radiology scans. We assumed that a model distracted by 17th-century French poetry or complex Python code would be inherently inferior at spotting a fracture. We wanted our AI to be as narrow as our own job descriptions.

So far at least, it turns out that machines’ intelligence is a horizontal force. The models that can reason through a legal brief or a thermodynamics problem are often better at diagnosing a patient because they understand the structure of logic itself. The Jack of all trades is no longer a master of none. In 2026, the generalist is becoming the master of everything, all at once, at a marginal cost of zero.

We still hunt for the moat in specialized datasets. We tell ourselves that proprietary data will save the niche players from the giants. We ignore the reality that general reasoning is cannibalizing the specialists before they can even finish their series B. The moat is not the data. It is the ability to think across boundaries.

This is the final habit we cannot kill: the belief that our own specific expertise still matters. We are specialized fund managers writing an article about the end of specialization. We are self-appointed experts in innovation who are regularly blindsided by the next Tuesday morning breakthrough.